THE PRACTICE OF INTEGRATING LEGUMES IN THE CROPPING SYSTEM: EVIDENCE OF IMPACT ON FOOD SECURITY AND NUTRITIONAL OUTCOMES OF SMALLHOLDER FARMERS IN UGANDA By Trhas Weldezghi A THESIS Submitted to Michigan State University in partial fulfillment of the requirement for the degree of Agricultural, Food, and Resource Economics Master of Science 2018 ABSTRACT THE PRACTICE OF INTEGRATING LEGUMES IN THE CROPPING SYSTEM: EVIDENCE OF IMPACT ON FOOD SECURITY AND NUTRITIONAL OUTCOMES OF SMALLHOLDER FARMERS IN UGANDA. By Trhas Weldezghi Legumes play an important role in a nutrition - focused agricultural strategy because they provide a myriad of environmental and nutritional benefits. To realize those benefits, legumes can be integrated as mono - cropping, intercropping, and rotation in a cropping system. The literature on the impact of legume - based cropping has been growing but not addressed in the case of Uganda. This study examines the im pact of legume - based practices on food security and nutritional outcomes of small - holder farmers using a nationally representative household survey for Uganda. A multi - step approach was used. In the first step, I assess the impact of different legume - based cropping on household level food security outcomes (i.e., calorie and protein produced, crop income, HDDS and MAHFP), and child level nutrition outcomes (i.e., extent and prevalence of stunting, underweight and wasting) along the agriculture - food security - nutrition impact pathway. As a second step, I attempt to identify the pathway through which legumes integration influences consumption and nutrition - related outcomes. The first step results suggest a positive and significant association of some legume - bas ed practices with production outcomes (e.g., robust results for legume - non - cereal rotation), and mixed or weak results for child nutrition outcomes. The impact on food consumption related outcomes (i.e., HDDS and MAHFP) remained insignificant in all cases except for legume non - cereal intercropping. In the second step, the study identified crop income as the main pathway to improve MAHFP and to reduce the prevalence of wasting, and production as the main pathway to increase HDDS. iii ACKNOWLE DGEMENT S Firstly, I would like to thank my major advisor Prof. Mywish Maredia from the Department dedication, mentorship, and unwavering support were invaluable . Thank you so much for being part of this journey and working tirelessly to make sure I successfully complete this thesis research. My sincere gratitude goes to the Mastercard Foundation Scholars Program for granting this opportunity to further my studie s. Without their support, this day would not have been possible, and for that I am thankful. I would also like to thank my thesis committee members, Dr. Nicole Mason and Prof. Irvin Widders, for their constructive feedback and guidance. I am grateful for their patience and unconditional support. I would also like to thank many graduate students in AFRE for their precious time and encouragement. Special thanks to Mukesh Ray; without his dedication and assistance, the data would have been twice as hard to c ode and decipher, and it would have been harder to get through this process. Last but not least I would like to express my profound gratitude to the Lord almighty for blessing me with this opportunity and rewriting my life. To my family, I cannot thank you enough; words are not enough to express my gratitude for your unfailing support. iv TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ........................ vi LIST OF FIGURES ................................ ................................ ................................ ................... viii CHAPTER 1. INTRODUCTION ................................ ................................ ................................ 1 1.1. B ackground and Motivation ................................ ................................ ................................ ................... 1 CHAPTER 2. LITERATURE REVIEW ................................ ................................ .................... 7 2.1. Introduction to the Concept of Food Security ................................ ................................ ............................ 7 2.2. Overview of Exist ing Literature and Methodological Approaches ................................ ............................ 10 2.3. Conceptualizing Legume - Based Cropping and Food Security Linkages ................................ .................... 15 2.4. Research Questions Addressed ................................ ................................ ................................ .............. 17 CHAPTER 3. DATA AND METHODS ................................ ................................ .................... 19 3.1. Data and Sampling ................................ ................................ ................................ ............................. 19 3.2. Treatment Variables ................................ ................................ ................................ ........................... 21 3.3. Outcome Variables ................................ ................................ ................................ ............................. 22 3.4. C ontrol Variables ................................ ................................ ................................ ............................... 26 3.5. Empirical Strategy ................................ ................................ ................................ .............................. 28 3.5.1. Step 1: Impact of Legume - Based Cropping on Household Food Security and Child Nutrition Outcomes ................................ ................................ ................................ ................................ ............................. 29 3.5.2. Step 2: Impact of Gross/Net Crop Income and Protein Produced on HDDS and MAHFP ............ 31 3.5.3. Step 2: Impact of Gross/Net Crop Income and Protein Produced on the Prevalence and Extent of Stunting, Underweight, and Wasting ................................ ................................ ................................ ....... 32 CHAPTER 4. RESULTS AND DISCUSSION ................................ ................................ ......... 33 4.1. Descriptive Analysis and Summary Statistics ................................ ................................ ........................ 33 4.2. Importance of Different Legume - Based Cropping in the Ugandan Context ................................ ............... 37 4.3. Impact of Legume - Based Crop ping on Household level Food Security and Child Level Nutrition Outcomes 41 4.4. Effects of Cereal - Legume Intercropping on Food Se curity and Nutrition Related Welfare Indicators ........... 42 4.5. Effects of Cereal - Legume Rotation on Food Security and Nutrition Related Welfare Indicators ................. 46 4.6. Effects of Legume - Noncereal Intercropping on Food Security and Nutrition Related Welfare Indica tors ...... 48 4.7. Effects of Legume - Non - Cereal Rotation on Food Security and Nutrition Related Welfare Indicators ......... 50 4.8. Effects of Legume Monocropping on Food Security and Nutrition Related Welfare Indicators .................... 53 4.9. Step 2: Effects of Gross/Net Crop Income and Protein Produced on HDDS and MAHFP .................... 55 4.10. Step 2: Effects of Gross/Net crop Income and Protein Produced on Child Nutrition Outcomes ............... 59 CHAPTER 5. CONCLUSION ................................ ................................ ................................ ... 63 APPENDICES ................................ ................................ ................................ ............................. 66 Appendix A: General Information on Legume Crops, Seasonality and Data Cleaning ................................ .... 67 Appendix B: Full Regression Results ................................ ................................ ................................ ........... 70 v REFERENCES ................................ ................................ ................................ ............................ 85 vi LIST OF TABLES Table 1: Summary Statistics for Whole Sample and for the Final Year (2013/14) ................................ ............. 35 Table 2: Prevalence of Stunting, Underweight and Wasting for Subsample of Households With 6 - 59 Months Children ................................ ................................ ................................ ................................ ......................... 37 Table 3 : Frequency of Legume - Based Cropping Adoption across the Four Survey Rounds . ................................ ... 39 Table 4 : Comparison of Means of all Dependent Variables, among Users and Non - users of Different Legume - Based Cropping ................................ ................................ ................................ ................................ ........................ 40 Table 5: Main Regression Results related to the Effects of Cereal - Legume Intercropping on Household Level Food Security and Child Level Nutrition Outcomes ................................ ................................ ................................ ... 45 Table 6: Main Regression Results related to the Effects of Cereal - Legume Rotation on Household Level Food Security and Child Level Nutrition Outcomes ................................ ................................ ................................ ... 47 Table 7: Main Regression Results related to the Effects of Legume - Noncereal Intercropping on Household level Food Security and Child Level Nutrition Outcomes ................................ ................................ ................................ ... 49 Table 8 : Main Regression Results related to the Effects of Legume - Noncereal Rotation on Household Level Food Security and Child Level Nutrition Outcomes ................................ ................................ ................................ ... 52 Table 9: Main Regression Results related to the Effects of Legume Monocropping on Household Level Food Security and Child Level Nutrition Outcomes ................................ ................................ ................................ ............... 54 Table 10 : Su mmary Regression Results for the Effects of Gross/Net Crop Income and Protein Produced on HDDS and MAHFP ................................ ................................ ................................ ................................ ................ 58 Table 11: Summary Regression Results for the Effects of Net crop Income, Calorie and Protein Produced on HDDS and MAHFP ................................ ................................ ................................ ................................ ................ 58 Table 12 : Summary Results for the Effects of Gross/Net crop Income and Protein Produced on Child Level Nutrition Outcomes ................................ ................................ ................................ ................................ ........ 61 Table 13: Summary Regression Results for the Effects of Net crop Income, Calorie and Protein Produced on Child Level Nutrition Outcomes ................................ ................................ ................................ ............................... 62 Table A 1: Summary on Panel data Formation and Data Cleaning ................................ ................................ .. 69 Table B 1: Fixed Effect Regression Results for the Effects of Legume - Based Cropping on Household Food Security Outcomes ................................ ................................ ................................ ................................ ....................... 70 vii Table B 2: Fixed Effect Regression Results for the Effects of Legume - Based Cropping on Continuous Child Nutrition Outcomes ................................ ................................ ................................ ................................ ........ 73 Table B 3: Fixed Effect Regression Results for the Effects of Legume - Based Cropping on Binary Child Nutrition Outcomes ................................ ................................ ................................ ................................ ....................... 75 Table B 4 : FE and POLS Regression Results for the Effects of Gross/Net Crop Income and Protein Production on HDDS and MAHFP ................................ ................................ ................................ ................................ ... 77 Table B 5 : Fixed Effect Regression Results for the Effects of Net Crop Income, Calorie and Protein Production on HDDS and MAHFP ................................ ................................ ................................ ................................ ... 79 Table B 6: Fixed Effect Regression Results for the Effects of Gross/ Net Crop Income and Protein Production on Continuous Child Nutrition Outcomes ................................ ................................ ................................ ............. 81 Table B 7: Fixed Effect Regression Results for the Effects of Gross/ Net Crop Income and Protein Production on Binary Child Nutrition Outcomes ................................ ................................ ................................ .................... 83 viii LIST OF FIGURES Figure 1: The Complex Nature of Food Security: Food Security Dimensions, Levels, and Components ................... 8 Figure 2: Conceptual Pathway of Legume - Based Cropping Impact on Food Security and Nutrition. ...................... 16 Fi gure 3 : Proportion of Households Using Different Legume - Based Cropping in Uganda ................................ ..... 38 Figure A 1: Examples of Different Types of Legume Crops ................................ ................................ ............... 67 Figure A 2: Yearly Calendar related to Planting and Harvest Months in Uganda ................................ .............. 68 1 CHAPTER 1. INTRODUCTION 1.1. Background and Motivation Food insecurity and malnutrition have been amongst the pressing problems facing many developing countries (FAO 2008; FAO 2017; Smith et al. 2006). Despite increased efforts towards alleviating food insecurity and malnutrition, progress towards reducing these problems has been below the desirable level (World Bank, 2007), especially in Africa and Asia. According to the study by Smith, Alderman and Aduayom (2006) on 12 Sub - Saharan African (SSA) countries, the prevalence of food insecurity was lowest in Uganda (37%) and highest in Ethiopia (76.4% ). A study by FAO indicates fluctuating trends in the prevalence of malnutrition in SSA; it steadily declined from 28.1% in 2000 to 20.6% in 2010 but saw an upward trend with an increase to 22.7% in 2016 (FAO, 2017). There are many strategies to address t he problem of food insecurity and malnutrition. These include improvements and investments in education (Lipton et al. 1998; Gaiha 1993), health system (Croppenstedt and Muller 2000) , economic growth and price stability (Timmer 2000) , climatic change, shock and conflict mitigation strategies (Wheeler and Von Braun 2013 ; Teodosijevic 2003; Schmidhuber and Tubiello 2007) , and political transformation ( Smith and Haddad 2015) . Agricultural productivity growth through improved technology and better crop management practices is also considered an important strategy towards alleviating food insecurity and malnutrition, especially since agriculture is the mainstay for the majority of the population, and an important contributor to national gross domestic product (GDP) in many developing countries (Wiebe 2003 , Godfray et al. 2010) . Even though many agricultural technologies help in boosting productivity, they are either costly or not readily accessible to smallholder farmers. As a result, farmers fail to maint ain productivity and are vulnerable to shocks in their resource base, the environment, and the economy 2 overall. Apart from that, low rate of technology adoption and inappropriate farming practices consequently trigger soil degradation, decrease crop produc tivity, food availability, and increases in food insecurity. According to the study by Ibrahim (2013), the percentage of Ugandan farmers using improved seeds was around 6%, whereas that of inorganic inputs was much lower during the past five years. The ado ption rate was especially low for farmers with less education and land size, and lacked access to credit, information, extension services and affordability. Among the myriad of crops grown by small - holder farmers in developing countries, legumes 1 ( see Appe ndix A ; Figure A 1 ) are one of the nutrient rich food crops that play an important role in agriculture. The ability to fix atmospheric nitrogen and supplement other non - legumes with mineral nitrogen make legumes essential components within various farming systems and are often promoted as part of a sustainable intensification strategy ( Messina 1999; Ncube et al. 2009 ; Giller 2001) . Legumes are important sources of protein, vitamins, micronutrients, a nd supplemental income. They also allow farmers to sell and consume seeds in green stages between harvests and can be stored in the dry stage after harvest without any loss of their nutritional content. Mor eover, legumes have high demand (i.e., part of the main diet in many developing countries) and marketability in many Sub Saharan African countries (Chianu et al., 2011) . In recognition of the multi - faceted benefits that pulses and legumes provide to farmers and consumers, the Foo d and Agricultural Organization of the United Nations had declared the year 2016 as the International Year of Pulses. There are many ways legumes can be integrated in the cropping system of smallholder farmers to realize their environmental and nutritional benefits. These include mono - cropping, 2 1 Legumes crops include lentils, peanuts, peas, beans, and other podded plants (Messina 1999) 2 Mono - cropping is the practice of planting a sole crop on the same land for a given growing season. 3 intercropping 3 , and rotation 4 (Anders, Potdar and Francis 1996; Manda et al. 2017) . These practices offer different types of advantages to smallholder farmers. Mono - cropping is often promoted as a strategy to increase the productivity of the crop itself, whereas inter - cropping and rotation are promoted more as strategies to increase the productivity, and sustainably intensify the whole cropping system. Continuous crop cultivation of any crop without fallows and input use deteriorates land quality and productivity. Crop rotation, especially with legumes, helps break the pest cycle and provides the benefits to the following crop from the residual nutrients left in the soil. Even tou gh intercropping may allow farmers to make intensive use of the limited land they have (Dwivedi et al. 2015) , help in improving soil fertility through nitrogen fixation, conserves soil due to land cove rage, reduces the need for complementary inputs like fertilizer, and allows farmers to have diversified means of income from growing multiple crops at one time (Kabunga, Dubois and Qaim 20 14; Fujita and Ogata 1992 ) , there are clear disadvantages of intercropping as it may also increas e competition for water and nutrients, can increase pest pressure, create difficulty in managing weeds and overall may result reduction in total productivity making management and selection of crops to be intercropped difficult ( Lithourgidis et al 2011; Thierfelder, Cheesman and Rusinamhodzi 2012) . Despite the potential for higher productivity of legume based practices, the impact of legume mono - cropping, rotation, and inter - cropping is highly variable, and depends on the soil and crop type (Ojiem et al., 2006) . These pr actices alone do not guarantee higher productivity. Their contribution to smallholder farming system is also subject to different socioeconomic and 3 Intercropping is the p ractice of growing more than one crop in a specific plot, and at a particular point in time. For a plot of land in a given growing season, the practice of intercropping and mono - cropping are mutually exclusive. 4 Crop rotation is growing different crops ea ch year/season over the same land. For a plot of land, the practices of rotation, intercropping and mono - cropping are not mutually exclusive over time. For example, a plot of land rotated with two different crops over two consecutive seasons could have bee n mono - cropped or inter - cropped in any given season. In other words, a given plot of land could be under the practice of rotation as well as intercropping and/or mon - cropping. 4 agroecological factors. Thus, understanding the effects of different ways of integrating legumes in the crop ping system, on household food security and nutrition outcomes within the socioeconomic context in which smallholders operate, is the focus of this study. I use Uganda as a case study to address this issue. Uganda was the first country to introduce legume cultivation in 1906 within the East African region (Byenkya 1988). According to the 2005/06 survey of 4.2 million agricultural households, the major crops grown in Uganda are maize (86.5 %), beans (80.8 %), cassava (74.3 %), banana (73.1 % ), and other cash and noncash crops (Uganda Bureau of Statistics 2007a, p. 46). Ugandan agriculture is dominated mostly by subsistence farming, scattered/fragmented small land size, low use of improved seeds, fertilizer, and pesticides ( FAO 2010 ) , and lack of extension services (BakamaNume 2010, p. 215). Uganda has a vast portion of arable land with diverse soil types. However, due to lack of appropriate conservation practices soil degradation has been a pressing problem (BakamaNume 2011; Olson and Berry 2003) . Over the past years, many studies have focused on understanding the linkages between agriculture and nutrition ( Manda et al. 2017; Magrini and Vigani 2016; Sauer et al. 2016; Kirk, Kilic and Carletto 2017; Azzarri et al. 2015; Kim, Mason and Snapp 2017) . A recent study by Sauer et al. (2016) assessed the impact of legume - based cropping on food security in Zambia. The study analyzed the effects of cereal - legume intercropping/rotation , and anyother legume - based cropping for the subample of cereal producing households.The stud y found strong effect (i.e., statistially significant and postive) effect of cereal - legume rotation, little or no statisitcally signifcant effect of cereal - legume intercropping, and postively significant effect of legume - other practices (i.e., legume inter cropping/rotation with non - cereal crops or legume mono - cropping) on different indicators of household welfare, including food security. Among the three legume - based cropping technologies they examined, the effect was much stronger in the case of cereal - leg ume rotation, while that of 5 cereal - legume intercropping and legume - other practices was not robust compared to cereal legume rotation. To the best of my knowledge, within the context of Uganda, the impact of such legume - based cropping on food security and n utritional outcomes of farming households has not been studied. Building on the analytical framework developed by Sauer et al. (2016), this study contributes to this gap by providing similar evidence on the impact of legume - based cropping on food security using a nationally representative dataset from Uganda. But it goes beyond the study by Sauer et al. (2016) or other previous studies by also examining the impact of legume - based cropping on nutrition outcomes. The goal of my research is to assess the pathw ays by which legumes can potentially impact household welfare and food security indicators, and whether these effects translate into nutritional outcomes for children. In this study, I use four waves of nationally representative Living Standard Measuremen t Survey (LSMS) data collected from smallholder farming households in Uganda. The panel nature of this dataset allows me to use one of the rigorous methodological approaches and techniques, namely the fixed effect model. My analyses focus on five types of outcome variables that are considered necessary conditions for achieving and measuring food security (Coates et al. 2013 ; Leroy et al. 2015) , namely crop income, calorie production, protein production, Household Dietary Diversity Score (HDDS), and Months of Adequate Household Food Provisions (MAHFP). In addition, I use anthropometric measurements of children below five years of age to assess the impact of the legume - based cropping on nutritional outcomes, namely stunting, wasting, and underweight, and examine the production and income, and food security effects on these child level nutritional outcomes. My thesis is organized as follow. In chapter 2, I present the review of literature on agriculture and food security linkages, existing methodological approaches pertinent to my research 6 topic, conceptual framework and res earch questions addressed. In chapter 3, I describe the data, sampling techniques, and specification of empirical strategy and models. In chapter 4, I discuss the results and main findings, followed by conclusions and recommendations in chapter 5. 7 CHAPTER 2. LITERATURE REVIEW The literature review is presented in four sections. The first section introduces the concept of food security, various definitions found in the literature and reviews literature on agriculture and food security linkages. The second section reviews the existing literature and methodological approaches used to assess the impact of agricultural practices/technologies in general, and legume based - practices/cropping in particular on food security and nutritio nal outcomes. The conceptual framework underlying the pathways from agriculture to food security and nutrition outcomes follows in the third section, and the last section explains the main research questions addressed. 2.1. Introduction to the Concept of F ood Security advanced over time (Maxwell 1996) . The earlier definition of food security had several drawbacks; at and nationally, food security retained the supply - side perspective of the overal l food system and had the same meaning as "self - sufficiency." According to that definition of food security, for any country to remain food secure, it was expected to produce all types of food required by its nationals. The standard definition of food secu rity which was stated during the World Food Summit in 1996 was that "Food security exists when all people, at all times, have physical and economic access to sufficient safe and nutritious food to meet their dietary needs and food preferences for a healthy and active life" (FAO, 1996). Unlike former definitions on food security, this definition emphasizes the nutritional composition and food safety besides the availability, access, and utilization of food to meet daily dietary requirements. Figure 1 illustr ates a framework developed by Leroy et al. (2015) in which food security is depicted in terms of four dimensions, namely -- availability, access, utilization, and stability. This framework further demonstrates that food security can be defined and studied at global, regional, 8 national, community, household and individual level. Component - wise, it represents quantity, quality, safety and cultural acceptability or preferences ( Leroy et al. 2015; Staatz et al. 2009 ) . Figure 1 : The Comple x Nature of Food Security: Food Security Dimensions, Levels, and Components Source ; Leroy et al. (2015) The relationship between agriculture, food security, and nutrition is not linear, and is the result of an interaction of many processes. Due to multidimensionality, it lacks robust conceptual/theoretical framework that depicts linkages from agriculture to food security (Pangaribowo, Gerber and Torero 2013; Haddad 2000) . Also, approaches that illustrate the causal pathways (either directly or indire ctly) through which agriculture impacts food security and nutrition varies depending on the level, dimension, and component of food security that is being addressed. There are many frameworks put forth and discussed in the literature that conceptualize the causal link between agriculture - food security - nutrition (Carletto, Zezza and Banerjee 2013) . Among these, the fram ework developed by Herforth and Harris (2014) focuses on household level linkages and illustrates how agriculture , through investment on new technologies or farming practices, impacts productivity and how the environment interacts through the existing practice and further 9 affects the nutrition and health conditions of farming households. In this framework, the major pathways comprise production, income, and women empowerment. Food production is among the main factors affecting the food security of farming households, especially in developing countries. However, availability of food from own production does not guaran tee access to food, required nutrients, and food security at all times (Carletto et al. 2013) . Moreover, its impact on nutritional outcomes depends on the quality, quantity, type, seasonality, and availability of food produced for consumption, which further depends on the existing market situation and each household's decision - making processes (Herforth and Harris 2014) . Besides, it is not very common for farming households to produce all types of food crops needed for a quality diet. Instead, they sometimes use their land to specialize in crop production, and prod uce surplus to generate income, which may allow households to access diversified diets. Thus, higher productivity per unit of land or other inputs used may contribute to food security and higher nutritional level through increased farm/crop income, assumin g there is access and availability of nutritious and affordable food in the market ( Herforth and Harris 2014; Manda et al. 2017; Haddad 2000) . Many studies have been conducted to examine whether higher agricultural productivity has linkages to increased income and dietary diversity of smallholder farmers (Pellegrini and Tasciotti 2014; Kirk, Kilic and Carletto 2017; Snapp and Fisher 2015; Sibhatu et al. 2015) . According to the effect of income on food consumption is higher for poor than wealthier households because of the higher income elasticity of food for poor p eople, particularly in terms of calorie and micronutrient consumption ( Deaton 2018 ; Skoufias, Tiwari and Zaman 2011) . Higher income may improve dietary intakes of farming households, not only through access to div ersified and nutritious healthy living. Moreover, evidence suggests that w omen with full control over resources are more 10 likely to allocate them better than men, and make pro - nutrition decisions for children (Haddad, Hoddinott, and Alderman 1997). Hence, as a third pathway may play a their own and productivity and income potential (Bhutta 2013; Herforth and Harris 2014). 2.2. Ove rview of Existing Literature and Methodological Approaches Technology adoption is an endogenous choice, which depends on a set of household and socio - economic factors, and it is difficult to disentangle its effect from any other household decisions or ex ternal factors. The literature on the impact of agricultural technologies on food security in SSA is not very extensive, although it has grown in recent years. Many studies have found positive impact of technology adoption on productivity and food securi ty (for e.g., Kassie et al., 2012; Kabunga et al. 2014; Magrini and Vigani 2016 ; Manda et al., 2017; Jaleta et al. 2018) , but they represent a specific context and are not generalizable. Moreover, due to lack of available data, many of these studies are unable to address all dimensions of food security ( Carletto et al. 2013). A study by Magrini and Vigani (2016) tried to look at the impact of technology adoption (i.e., inorganic fertilizer and improved maize seed) on food security in Tanzania, using one year cross - sectional data from Tanzania National Panel Survey data series . The f ocus of the study was assessing the effect on four pillars of food security, namely availability, access, utilization and stability. The study addressed the issue of selection bias by using propensity score matching (PSM). The PSM method quantifies the imp act of adoption based on similar observable covariates of adopters and non - adopters. To control selection bias and unobserved heterogeneity, and as a robustness check, endogenous switching regression model was used. However, due to the cross - sectional natu re of the data, the study was unable to capture the long - term effects of adoption on food security. Similarly, Jaleta et al. (2018) studied the impact of imp roved maize varieties on food 11 security by using nationally representative cross - sectional data for Ethiopia. The study also used endogenous switching regression model as means to control endogeneity and selection bias. Manda et al. (2017) looked at the ex - ante effects of maize - soybean rotation on household food security using experimental and observational data from household sur veys conducted by the Institute of Tropical Agriculture (IITA) and the International Maize and Wheat Improvement Centre (CIMMYT), respectively. The study assessed the impact of maize - soybean rotation on changes in household income and poverty alleviation u sing market level economic surplus approaches. To estimate surplus changes on individual households, as opposed to the changes in market surplus, the study incorporated household level analysis along with the market level information. Their results suggest ed a positive impact of maize - soybean rotation on income and poverty reduction of small - scale farmers compared to monocropping. A limitation of the data, which was cross - sectional, was limited control over unobservable factors of the analysis. Methods bas ed on panel data are more advantageous than cross - sectional data and enable researchers to capture the heterogeneous effect of the existing trends within a specific context. In the last decade or so, the use of panel data techniques in the impact evaluatio n literature has increased steadily. Moreover, researchers have been developing panel data models that are compatible with specific data and variables of research interest, which has led to consistency and efficiency in estimation. Some recent studies focu sing on agriculture - nutrition linkages that have used panel data include: Azzarri et al. 2015, Sauer et al. 2016, Kirk et al. 2017, and Kim, Mason and Snapp 2017 . My research builds on the panel data methodologies of some of these recent studies, which are described below. The study by Sauer et al. (2016 ) looked at the impact of legume - based technologies on household welfare using the Rural Agricultural Livelihoods Survey ( RALS) panel data of two waves for Zambia. The motivation behind the study was to assess the production and income pathways 12 through which legume - based technologies, such as intercropping and rotation impact food securi ty among cereal growing households. The outcome variables used reflect different indicators along the pathway from production of calories and proteins to crop income, Household Dietary Diversity Score, and Months of Adequate Household Food Provision. Due t o the differences in the variable type (continuous vs. count) amongst these outcome indicators, the study used different models, namely household fixed effect model, correlated random effects negative binomial, and pooled ordinary least squares models. The study also used two stage least square instrumental variables model as a control for self - selection bias and endogeneity problem. The study found a positive and significant impact of cereal - legume rotation on household welfare, little effect of cereal - leg ume intercropping, and mixed effect of other legume - based technologies; mono - cropping, rotation/intercropping with non - cereal crops on household welfare indicators. Adequate animal product consumption, associated with livestock ownership, can potentially impact the nutritional status of farming households and children. Due to high prevalence of livestock ownership but higher stunting rate among children in Uganda, Azzarri et al. (2015) tried to look nutrition outcomes of children living in rural areas, using two waves of Living Standard Measurement Survey (LSMS) panel data for Uganda. The study identified a subs tantial difference in ASF consumption between livestock owners and non - owners; owners tend to consume higher amount of ASF than non - owners. It further indicated (i)a significant effect of large ruminant ownership on dairy food consumption but not on beef c onsumption, (ii) a non - significant effect of small ruminant ownership on sheep and goat meat consumption, but positive effect of poultry ownership on chicken consumption. In translating the positive effect of ASF consumption on child nutrition outcomes, th e study found a weak correlation between livestock ownership and child nutrition outcomes (underweight and wasting) and no association with stunting. However, the 13 results were sensitive to the specific age groups of children; with higher effect on older ch ildren (children between ages of 24 to 59 months). The study did not find any effect of ASF consumption on nutrition outcomes of older age groups like lactating mothers and women at reproductive age and called for further research on the impact of other co nsumption components that affect nutrition. Higher income is often associated with better nutritional outcomes. However, does the source of income influence the nutritional outcome? The study by Kirk et al. (2017) addressed this question by estimating the effect of different sources of income on short - term child nutritio n outcomes. The study examined the effects of crop and non - crop income; different sources of income within agriculture, and the overall income of the household regardless of the source. The study used three waves of LSMS panel data for Uganda and empirical ly analyzed the data using child fixed effect model with and without controlling for child - level characteristics. Specific to Uganda, the impact of agricultural income nutrition outcomes was negative indicating (i) a negative and significant impact of the share of crop income and crop consumption on the height - for - age z - scores (HAZ), implying low nutrient crops production and own crop consumption, (ii) non - significant impact of the shares of livestock and wage income and conversely, positive and highly sign ificant effect of self - employment income signaling higher correlation with child nutrition outcomes. Due to lack of convincing and strong instrumental variables for income, the study findings are not conclusive on the causality of income sources on child n utrition outcomes. Households usually adopt more than one technology, but most studies in the literature has focused on assessing the impact of a single technology. Single - technology studies do not capture the differential impact of technologies per se. Wi th main intention of filling this gap on literature, the study by Kim, Mason, and Snapp (2017) looked at the differential impact of sustainable farming practices (SFM) on child nutrition outcomes within a framework if multiple technology adoption 14 decision. To capture the e ffect of unobserved heterogeneity and selection bias, the study used multinomial endogenous switching regression model. The study finding indicated positive effect of all treatment practices on height for age z - scores (HAZ), but only sustainable intensific ation had a positive effect on weight for age z - scores (WAZ). Overall, the study found a positive effect of sustainable farming practices on child nutrition outcomes. The review of literature presented in this section highlights many factors that can affec t food security and nutrition outcomes namely (i) agricultural technologies embedded in input use such as fertilizers and improved seeds, (ii) animal source products, (iii) income and different sources of income, and (iv) sustainable farming practices such as intercropping or rotating cereal or non - cereal crops with legume crops. However, these studies represent specific contexts and situations, and are not generalizable. Motivated by the importance of legume crops in Uganda, this research stu dies the impact of legume - based cropping on household food security and household level nutrition outcomes of small - holder farming households in Uganda. My study contributes to the literature in two ways. First, by looking at the effects of all the ways le gumes can be integrated in cropping systems i.e., monocropping, intercropping with cereal and non - cereal crops, and rotating with cereal and non - cereal crops. Second, it extends the analysis to include the effects of legume - based cropping on child nutritio nal outcomes, and examining how legume - based cropping impact on production and income outcomes translate to child nutrition outcomes. Moreover, to the best of my knowledge, there is no study done on the impact of legume - based cropping in Uganda, hence this study also contributes to country - specific literature on the impact of legume - based cropping on indicators of food security and nutritional outcomes along the agriculture - nutrition linkages pathway. 15 2.3. Conceptualizing Legume - Based Cropping and Food Secu rity Linkages Improved agricultural technologies are considered necessary for attaining higher productivity, and many countries have experienced structural transformation via adoption of agricultural technologies. Having said that, higher productivity via agricultural technology does not guarantee higher nutritional level at all times. Moreover, many of the existing agricultural technologies are not easily accessible and affordable for smallholder farming households, leading to low adoption rates (Lunze et al. 2012) . For a nutrition - focused agricu ltural strategy, legumes are good alternatives, especially for small - scale farmers in developing countries. In the context of Sub - Saharan Africa, legumes are widely grown and are easily adaptable at small scale production. As mentioned earlier, intercroppi ng/rotating legumes with cereals or any other crop has a multitude of benefits and can potentially impact food security and nutrition of farming households through production, income, Figure 2 illustrates these pathways th rough which legume - based cropping such as rotation, intercropping, and mono - cropping are conceptualized to impact household food security and nutrition. In the production pathway, the biological characteristics of legume crops such as the ability to fix ni trogen and enhance soil quality come into play and can induce higher total productivity of the cropping system compared to systems without such legume integration (Giller, 2001; Hartwig and Ammon, 2002; Manda et al., 2017) . 16 Figure 2 : Conceptual Pathway of Legume - Based Cropping Impact on Food Security and Nutrition. Source; Adapted based on Sauer et al. (2016) As shown in Figure 2, households that adopt legume - based practices have a higher potential for increased productivity, given other conditions are constant. Increased productivity further might lead to more production of calories, protein, and other micro - n utrients, which implies increased food availability for the household. It also implies increased income from sales of surplus harvest. diversified diets. Sim ilarly, higher crop income may also increase household spending on sanitation and health products/services, which can interact positively with food consumption to enhance es, leading to Awareness on t he Benefits of Legume - Based Practices Increased Crop Productivity Adoption of Legume - Based Practices (Intercropping/ Rotating a nd Monocropping Legumes with Cereals/any o ther Crop s Increased Income Higher Calorie and Protein P roduction Increased Expenses o n Diversified Food Increased Spending o n Healthcare a nd Sanitation Adequate and Stable Calorie , Protein, and Micronutrient intake Increased Micro - Nutrient Absorption 17 better nutritional outcomes. Even though empowered mother's role is vital in the proper implementation of infant and young child feeding practices, and the overall consumption and expenditure decisions within the household, due to lack of da ta this study did not look at the third In summary, in this study the biological, nutritional and environmental benefits of legume - based cropping and its interaction with the socio - economic setting of small - scale farmers in developing countries are hypothesized to impact the household welfare, namely food security and nutrition, through the production, and income. Hence, I try to test the hypothesis on whether legume - based practices impact food security and nutrition outc omes, within the context of Uganda, identify potential pathways and address the following research questions 2.4. Research Questions Addressed In the context of the food security conceptual framework depicted in figure 1, the focus of my research is on the quantity and quality components of food security at the household and individual levels. The indicators I use in my analysis reflect the availability, access, and utilization dimensions of food security, and on nutritional status outcomes with a focus on children less than five years of age (more discussion on these indicators is in the following chapter). These indicators fall across different nodes of the impact pathway depicted in Figure 2. As mentioned earlier, legumes play an important role in bot h the production and dietary systems of Ugandan farming households. In the context of Uganda, my research addresses following research questions: 1. Do legume - based practices i.e., cereal - legume rotation, cereal - legume intercropping, legume mono - cropping, and other ways of integrating legumes in the cropping system, impact food security and nutritional outcomes of farming households in Uganda? Do these impacts vary across these practices? 18 2. What are the main pathways through which agriculture is linked to food s ecurity outcomes within the Ugandan smallholder farming context? 3. Do the production and income effects of legume - based cropping (if any) translate to improvements in child nutritional status as reflected in the prevalence of child level malnutrition indicat ors? 19 CHAPTER 3. DATA AND METHODS 3.1. Data and Sampling This study uses four waves of the Living Standards Measurement Study - Integrated Surveys on Agriculture (LSMS - ISA) data for Uganda, which took place between years 2009 and 2014. Out of 783 e numeration areas (EAs) across the country, 322 were selected for the Uganda National Panel Survey (UNPS) in 2009/10. The EAs within each stratum were selected with implicit stratification as urban/rural and district with equal probability. The surveyed sam ple of the population includes urban/rural residents in Central without Kampala, Eastern, Northern, and Western regions. The survey used two - stage stratified random sampling and is a nationally representative panel survey for Uganda. The survey was conduct ed with structured questionnaires administered at the household and community level. For households that actively engaged in agriculture, the survey includes agricultural modules. The household questionnaire includes information on general household demogr aphic characteristics, consumption, expenditures, assets owned, educational background, financial and transport services, and anthropometrics measurement for children 6 - 59 months. The agriculture questionnaire includes landholding owned and rented, inputs used, and crops grown and sold by each household. The community questionnaire inquiries about service availability in the community, education, health services, tra nsport and work infrastructure. The agriculture module was conducted twice a year to account for seasonality in p roduction, consumption, and marketing . The first agricultural season is from January to June and the second agricultural season goes from July to December (Interviewer manual, 2009). The harvesting months for both cereals and legumes i s from May to August, in the first season, and September to January, in the second season (see Appendix A: Figure A 2) ( FAO, 2010) . 20 Appendix A: Table A 1 shows the brief summar y on the number of househo lds surveyed, observations pre and post data cleaning, and number of observations across the four panels (waves) of datasets used in this study. Column 6 of Appendix A; Table A 1 represents the number of households that have full information on both agricu lture and h ousehold questionnaires. Across all four waves, the unbalanced panel includes 9,018 observations. Out of these observations (N=9,018), 14.07%, appeared in one round only, 7.85% in two rounds, 29.24% in three rounds and 48.84% in all four rounds. 5 For the panel data models used in this paper, 14.07% of households that appeared only in one round were dropped from the analysis. Furthermore, to examine the effects of crop rotation with legumes in a given agricultural year, the analysis was restricted to those households that grew crops in both the seasons in a year. The analytical sample finally used in this study includes an unbalanced panel of 6,489 observations across 2097 households, which represents about 72% of total number of observations acros s all four rounds. Due to the large loss of observations over the four survey rounds, I conducted an attrition test using the method outlined in Wooldridge (2002; p. 585) . I created a new attrition variable (att t+1 ), that takes the value 1 , if a household appears in at least two rounds and grew crops in two se asons in a year, and taking the value zero otherwise , relative to a given year. Using the overall unbalanced panel, I regressed all dependent variables on all explanatory variables and the attrition variable (att t+1 ) using the fixed effect model. Based on that test, I failed to reject the null hypothesis of no attrition bias in all outcome variables. The coefficient on the attrition variable was non - significant in all of the outcome variables with p value ranging from 0.21 to 0.69. The study further incorpo rated child anthropometrics measurement data and tried to look at the impact legume - based practices on the prevalence of stunting, underweight, and wasting among 5 Using 2009 as the base sample, the attrition rate was minimal in 2010/11 and 2011/12, but 2013/14 only 67.4 % of households were among those that were interviewed in the previous wave (2011/12), while 32.6% were rotated. 21 children. For this analysis, I use a subsample of households that have children between ages o f 6 and 59 months. Moreover, for my analysis I only included households that have child level anthropometrics data in at least two survey rounds and grew crops in two seasons in a given agricultural year. In total, this analysis on child level nutritional outcomes is based on an unbalanced panel of 3,490 children aged 6 to 59 months across 922 households. 3.2. Treatment Variables Following Sauer et al (2016), this study used five main treatment variables and presented each treatment variable as a (1) dummy/ binary; taking the value one if the household is a user of the given legume - based cropping and zero otherwise; and as a (2) continuous variable representing the area planted under each type of legume - based cropping. The binary treatment includes cereal - leg ume intercropping, cereal - legume rotation, legume non - cereal intercropping, legume non - cereal rotation, and legume monocropping. Similarly, the continuous treatment variables represent total area planted to each type of legume - based cropping. The binary tr eatment variable captures the prevalence of each practice, while the continuous treatment variable captures the extent of use of each practice. - based Note that these are all household level decisions and thus not mutually exclusive treatment variables. In other words, a legume growing household could potentially be integrating legumes in the cropping system in one or multiple ways in any given year (see the explanation below on the definition of treatment variables). In this study, I created the five - treatment variables using crop level data of the Agriculture maize, and legumes and includes perennial crops, 22 - - comprises at least one non - cereal or non - include any of the legume crops. In the next step, I identified each plot as - defined each plot as (1) cereal - legume intercropping, if the plot consisted of both cere als and legumes in one season; (2) cereal - legume rotation if the plot consists of cereals the first season and legume in t he second season, and vice versa; (3) legume non - cereal intercropping if the plot consists of legumes and other crops in the same season; (4) legume non - cereal rotation if the plot consists of legume plot in the first season and other crops in the second s eason, or vice vers a, and (5) legume - monocropping if the plot consists of only legume crop in a year. Lastly, I categorized each household as a user or non - user of the five legume - based cropping practices if the household practiced them in at least one of its plots. 3.3. Outcome Variables According to the definition noted earlier (and as indicated in Figure 1), food security incorporates four dimensions -- namely, availability, access, utilization, and stability. Household food access refers to the adequacy in quality and quantity of dietary requirements for a productive life (Swindale and Bilinsky 2007), whereas utilization refers to the biological capacity and making the best use of available food for productive and healthy life. Indicators of nutr itional status such as anthropometric measures for children reflects the utilization component of food security. In assessing the pathways through which agriculture impacts food security and nutrition, this study includes food security outcome indicators d erived either directly from the dataset or by calculating them. The outcome variables used include calorie produced per capita per day, protein produced per 23 capita per day, gross/net crop income, weekly HDDS (Household Dietary Diversity Score), and MAHFP ( Months of Adequate Household Food provisions). All of these indicators measure one or multiple dimensions of food security. For example, calorie and protein produced per capita per day represent food availability, gross and net crop income represents both food availability and food access, HDDS represents food access dimension, and MAHFP represents both food access and food stability. Also, based on anthropometrics measures, I also used indicators of nutritional status of children based on the z - scores and whether a child was stunted, underweight or wasted, measuring the utilization dimension of food security. Naturally, all crops have diversified nutritional content, and to capture the impact of legume - based cropping on food security and nutrition, it is in appropriate to use the total amount of production of different crops in aggregate as an indicator of food availability. As such, researchers have designed means whereby all crops can be converted to a standard metric; converting total production to calori e and protein produced. To calculate this, I first converted the total crop production harvested into a standard and common unit (Kg). After converting all crops into a standard unit (Kg), the total kilograms of each crop were converted to their equivalent calorie (calories) and protein (grams) content. The study used crop calorie data conversion table and google conversion rate of major crops 6 . Even though economic growth/increased income per capita has a potential to alleviate food insecurity, metrics on the food access node of agriculture - food security pathway are weakly correlated to increased income, while those on the food utilization node are highly correlated to increased income (Tandon et al. 2017). To capture that, I included gross and net crop in come representing the availability and access dimensions of food security. The gross crop income 6 Available at; http://iopscience.iop.org/17489326/8/3/034015/media/erl472821suppdata.pdf . 24 represents the monetary value of all the crops grown by each household; i.e., the gross value of crop product ion (i.e., quantity harvested multiplied by crop p rice), while the net crop income is gross income less the costs of purchased seed, organic/inorganic inputs, pesticides, transportation, and hired labor. To account for inflationary measurement bias, I adjusted the gross/net crop income to 2009 market valu e using the composite consumer price index for food crops in Uganda 7 . Similarly, all other monetary variables used in the analysis were converted to the 2009 monetary value. The net crop income might not necessarily correlate with the food access path of the food security but might induce consumption of different quality foods, reflecting higher correlation to the dietary intake node of the food security pathway (recall Figure 2). HDDS is a frequently used measure of food access component of food security, and measures the quality of food consumed at the household level. According to the Food and Nutriti on Technical Assistance (FANTA) 8 for HDDS measurement guideline, HHDS is measured based on 12 food groups/categories; 1. Cereals, 2. Root and tubers, 3. Vege tables, 4. Fruits, 5. Meat, poultry & offal, 6. Eggs, 7. Fish and seafood, 8. Pulses/legumes/nuts, 9. Milk and milk products, 10. Oil/fats, 11. Sug ar/honey, and 12. Miscellaneous. The standard HDDS indicator is a count variable ranging from 0 to 12, repres enting the number of food groups/categories (out of 12 categories) consumed by each household in the past one day, reflecting the variety of foods consumed by the household. In the LSMS dataset for Uganda, each household was asked about their food consumpt ion behavior in the past seven days rather than past one day. Thus, in the absence of data on foods consumed in the past 24 hours, which is used to estimate the standard HDDS, I use the weekly HDDS based on the past one week recall period. Many previous st udies have used such weekly HDDS indicator as a 7 Available at ; https://www.ubos.org/onlinefiles/uploads/ubos/cpi/junecpi2011/June_2011_CPI.pdf , pp 6 ; https://www.ubos.org/onlin efiles/uploads/ubos/cpi/cpiMarch2015/FINAL%20CPI%20Release%20 - March%202015.pdf ; pp 5). 8 Available at; https://www.fantaproject.org/sites/default/files/resources/HDDS_v2_Sep06_0.pdf . 25 measure of access to diet quality (Kibrom and Qaim 2016, Snapp and Fisher, 2015). Similar to these previous studies, I use a weekly HDDS indicator, as opposed to the standard daily HDDS indicator. The MAHFP is a count variable asking households to report on the number of months they indicator variable is the difference between 12 and the number of months household wa s food insecure. Usually, MAHFP is considered an indicator of food access. However, given the fact that it captures food access over the past 12 months, in a way it is also a measure of food stability. In other words, a household who has access to food all 12 months (MAHFP=12) prior to the interview time, and also implies more stability in food security over that time period, and vice versa. According to Swindale and Bilinsky (2007), MAHFP data should be collected during the time before harvest to avoid the bias of recall time. For the household survey I used, the data was collected throughout the year and there could be potential b ias as farmers might report the state of food security in the interview date rather that year - round estimates (i.e., recall bias could be an issue as well) . Also, the survey did not collect information on MAHFP in the second wave (2010/11). Thus, the analysis in this paper for this outcome indicator is based only on three waves of panel data (2009/10, 2011/12, and 2013/14). The stu dy included two child nutrition outcome indicators, namely (1) stunting 9 , which is an indicator of chronic (long - term) malnutrition, (2) wasting 10 , which is an indicator of acute malnutrition as a result of chronic disease or starvation; it is an indicator of current and short - term nutritional status of children (WHO 2010), and (3) underweight 11 , which is a composite measure of stunting or wasting or both, and thus complex to interpret. These indicators were created using child level data for children aged 6 - 59 months. In doing so, I First created child level HAZ (height for 9 If Height for Age (HAZ) is below 2 standard deviations from the mean/median 10 If Weight for Height (WHZ) is below 2 standard deviations from the mean/median 11 If Weight for Age (WAZ) /Weight for Height (WHZ) or both are below 2 standard deviations from th e mean/median 26 age), WAZ (weight for age), and WHZ (Weight for height) z - scores 12 using the g uidelines from 2006 WHO child growth standards; zscore06 Stata command (Leroy 2011). After calculating the z scores for HAZ, WAZ and WHZ, I defined each child as wasted/underweight/stunted, based on the z - score values. A child with a HAZ/WAZ/WHZ z - score of below negative two standard deviations ( - 2 SD) from the reference median/mean is categorized as stunted/underweight/wasted. Finally, I generated a dummy and count type variables for stunting, underweight, and wasting at the child level. In the case of the binary child nutrition status outcomes, the value 1 indicates stunting, underweight, and wasting and zero otherwise. 3.4. Control Variables As in many developing countries, farm households in Uganda act as both producers and consumers of own production. I n other words, choices regarding production (including technologies/practices to use for production) and consumption decisions are non - separable. Moreover, farmers face incomplete markets and act as utility maximizers (de Janvry, Fafchamps and Sadoulet, 19 91), as opposed to profit maximizers. Hence a non - separable household model is ideal in the context of Uganda. In addition, adoption of legume - based cropping is correlated to production, consumption, and market - related variables. In the study by Manda et a l. (2017) age of household head, land ownership, and education of household head are among the main factors that affect adoption decisions, and therefore I included them as control variables. The succeeding paragraphs discuss the details of production, con sumption and market - related variables included in my models. Production - related control variables include (1) plot level characteristics such as the total number of plots owned, total area planted/owned; (2) household - level characteristics affecting 12 Z - score (or SD - score) = (observed value - median value of the reference population)/standard deviation value of reference (Source; http://www.who.i nt/nutgrowthdb/about/introduction/en/index4.html 27 produ ction include total value of household assets, total livestock units 13 , other sources of income, distance to the roads, distance to the market; (3) district - level variables such as prices of main crops like beans, maize, and groundnut right before planting season. As farmers planting decisions primary depend on the crops before the planting season, I took the previous year and last season prices in each of survey rounds. Also, household - related characteristics that affect both production and consumption incl ude age, gender and educational level of the household head, and household size. Community - level variables include average temperatures and annual rainfall. To account for the impact of access to agricultural and market information, I include a dummy var iable equal to one for radio or mobile ownership or equal to zero for those who do not own. A year dummy was included to account for time differences across the four survey rounds. To further differentiate how legume - based cropping affect households who r eside predominantly in rural or a peri - urban setting, of the legume - based cropping. To take account of time and location - specific variability, I included an interaction term for region and year. For the child nutrition status outcome analysis, I included many of the (i) household - level variables listed in earlier paragraph and additional variables such as number of children between ages 6 to 59 months; a nd (ii) child level variables such as age of child in months completed, gender of child taking the value 1 for male and 0 for female, and a dummy for diarrhea, taking the value 1 for a child having diarrhea two weeks before the survey period and 0 otherwis e. Moreover, since nutrition status is closely linked with access to quality water, sanitation and hygiene (i.e., WASH indicators), I included (iii) WASH indicator variables that take a value of 1 if household has access to safe drinking water, flush toile t or ventilated improved covered pit latrines or zero otherwise. The study 13 The total livestock owned by the household was converted in to common units using total livestock unit conversion factor for sub - Saharan Africa (Njuki et al. 2011) . 28 further includes (iv) community - level variables namely, access to government - owned clinics and access to a market, both taking the value of one if the household has access, and 0 ot herwise. 3.5. Empirical Strategy As mentioned earlier, legume - based cropping is hypothesized to play a prominent role in the agriculture - food security - nutrition nexus. The fundamental objective of this study is to empirically test this hypothesis about wh ether and which legume - based cropping impacts food security (as measured by indicators reflecting food availability, access, and stability), and through which pathway production or income. Whether and how a household integrates legumes in their cropping sy stem is equivalent to a technology adoption decision. Technology adoption is an endogenous choice that households make and farmers might self - select whether to adopt or not. Hence there are many other unobserved factors that might affect the adoption deci sion and the outcome variables as well, leading to potential endogeneity problems. To address that, the study looked for variables that could be potential instruments (i.e., proportion of households using each legume - based practice excluding the household under consideration, access to agricultural extension services, and rainfall data. Nonetheless the f test on each instrument based on the reduced form equation was not strong enough. Hence the study was not able to address the potential endogeneity problem s. As noted in Chapter 2, In the context of a non - separable household model, my empirical analysis focuses on estimating the impact of the five types of legume - based cropping/practices on production outcome variables (calorie and protein produced), agricul tural income (gross/net crop income), consumption outcome variables (HDDS, MAHFP), and nutrition outcome variables ( extent and prevalence of stunting, underweight and wasting). The treatment variables used include cereal - legume intercropping, cereal - legume rotation, legume non - cereal intercropping, legume non - cereal rotation and legume monocropping. I take a two - step approach in my empirical strategy to 29 address my research questions. First, I examine the direct effects of different legume - based cropping on all the outcome variables, irrespective of where they fall along the agriculture - food security - nutrition impact pathway illustrated in Figure 2. In the second step 14 , I try to assess the pathway (production, income, or both) through which consumption relate d outcomes are impacted by regressing consumption outcome variables on production and income outcome variables. Lastly, I also analyze if increased crop income and protein/calorie production have any effect in reducing the prevalence of or extent of stunti ng, underweight and wasting of children within each household (i.e., nutrition outcome). To correct for serial correlation and heteroscedasticity, I clustered the standard errors at household level. The details of each step are as described below. 3.5.1. Step 1: Impact of Legume - Based Cropping on Household Food Security and Child Nutrition Outcomes In assessing the impact of different legume - based cropping on food security outcomes, due to high attrition rate in the case of using balanced panel, the study analysis was restricted to unbalanced panel, where attrition bias was not a problem. Even though there is a difference in data type, distribution of the outcome variables , and existing tradeoffs regarding efficiency and consistency among different panel da ta models, I used linear models such as Pooled Ordinary Least Square s (POLS) and Fixed Effect (FE) models as oppose d to Correlated Random Effects ( CRE) m odels, which can potentially create bias in the case of unbalanced panels. I first start with the simp le pooled ordinary least square s (POLS) regression model. The main drawback of POLS model is that it does not allow controlling for individual - specific and time - invariant unobserved heterogeneity. As argued earlier, legume technology adoption is an endogen ous choice that can be influenced by many unobservable factors (e.g., inherent management ability, risk attitudes, health status, motivation, etc.) that can potentially also affect household welfare and 14 Note that the realization of this step depends on the results of step 1. In other words, it is contingent upon finding significant effect in step 1 of treatments on the production and income outcomes. 30 nutrition outcomes. Thus, using POLS can potentially result in biased estimates. However, as the assumptions of POLS are less restrictive and weaker than any other panel data models, I report POLS regression results as preliminary results and for comparison and robustness check as well. To control for time c onstant unobserved heterogeneity, the fixed effects model is ideal and widely used for panel data analysis. By allowing the un - observables to be correlated with observable covariates, it reduces the time invariant unobserved heterogeneity problem. Equation 1 below represents household level fixed effect model for the household level food security analysis. 1 Where represents food security outcome variables for each household h, at time t, represents the constant term, denotes legume technology variables in discrete form (dummy) and continuous form. The discrete form of takes the value one and zero fo r adoption and non - adoption, respectively. While, the continuous form of represents area of land planted (acres) under each legume - based cropping for each household h, at time t . represents a set of household characteristics including total l and size planted, number of plots, total value of household assets, total livestock units owned, a dummy for non - farm income and residence. It also includes a set of community and market information characteristics; a dummy variable for mobile and radio ow nership, and average prices of the major crops from previous seasons. The household fixed effect, , year dummy , and the idiosyncratic errors are among the major variables in the model. The coefficients of interest in this model are , ref lecting the impact of each legume - based practice on the outcome variables. A positive and significant , implies a positive impact of legume - based cropping on food security outcome variables. For the child level nutrition outcomes analysis, the fixed ef fect model/Linear Probability Model (LPM) is as shown in Equation 2. 31 Where represents nutrition out come for child c , in household h , and at a time t, (i.e., child nutrition outcomes are in both continuous vs dummy forms), represents the constant term, denotes legume technology variables in discrete form (dummy) and continuous form, represents a set of household charact eristics, represents a set of child level characteristics including age of a child, and a dummy for child gender and diarrhea. The , , and represent the household fixed effect, year dummy and the idiosyncratic errors respectively. The coefficients of interest in this model are , reflecting the impact of each legume - based practice on the child level nutrition outcome variables. A negative and significant coefficient in , implies a positive association/ impact of legume - based techn ologies on binary type child nutrition outcomes (1.e., stunting, underweight and wasting) and the reverse for the continuous type of child nutrition outcomes (i.e., HAZ, WAZ, and WHZ z - scores). 3.5.2. Step 2: Impact of Gross/Net Crop Income and Protein Pr oduced on HDDS and MAHFP As mentioned above the first step was to assess the impact of legume - based cropping on household welfare, but that does not tell the specific pathway through which the consumption level outcomes are impacted, since these outcomes a re further down the impact pathway of the agriculture - food security linkages (see Figure 2). After regressions of production outcome variables on legume - based practices, if I observe a positive and significant coefficient on net crop income and calorie/pro tein per capita, then it represents a positive impact of legume - based cropping on production outcomes (the first node in the impact pathway), and it would warrant this second step, where I analyze whether the impact of legume - based cropping on production o utcomes translates into consumption outcomes. This was done by regressing HDDS and MAHFP, which are the two indicators of food consumption related outcomes, on gross/net crop income and protein 32 production 15 . In this step, a positive and significant coefficient estimates on any of the production outcomes would represent the specific pathway. This second step is crucial as it unpacks the heterogeneous effects (disaggregates each production outcome effect) on nutrition outcomes or the effect s of other unobserved covariates that can potentially affect HDDS and MAHFP, aside from net crop income and protein produced. 3.5.3. Step 2: Impact of Gross/Net Crop Income and Protein Produced on the Prevalence and Extent of Stunting, Underweight, and Wa sting Another dimension that the study looked at was, examining whether legume - based cropping further improves malnutrition status of children (i.e., stunting, underweight, and wasting). In doing this, the study used dummy and continuous variables for st unting, underweight, and wasting. For continuous type dependent variables, I used POLS and FE models for the continuous type , and Linear Probability Model (LPM) for the dummy type depe ndent variables as these models as LPM with FE model control time invar i ant unobserved heterogeneity and requires fewer distributional assumptions . Similar to the second step on household food security analysis, this step is done by regressing child level nutrition outcomes on protein and gross/net crop income . A negative and significant coefficient on the protein and gross/crop income indicates positive correlation with the binary type child nutrition outcomes (i.e., negative association with the continuous type child nutrition outcomes), and pathway through which legume - based practices can potentially affect child level nutrition outcomes. 15 Protein production was selected as treatment variable mainly due to more protein content available in legumes than calories. Also, with the high correlation between protein and calorie produced (0.87), it is not possible to use both as main treatment variables in a model. 33 CHAPTER 4. RESULTS AND DISCUSSION In this chapter I first present the summary statistics and descriptive analysis of all the outcome, treatment, and control variables used in the analysis. Then I present the results of econometric analysis, and discuss the main findings related to the impa ct of legume - based cropping on the food se curity and nutrition outcomes. 4.1. Descriptive Analysis and Summary Statistics Table 1 shows descriptive statistics of all the relevant variables included in the analysis. On average, a typical farming household in Uganda has a family size of six members, a land size of four acres, owns 3.5 plots, has total livestock units of 2, and a household head who is 48 years of age and has 5 years of education. The urban farming household comprises around 1 2 %, while the rural households account for 8 8 % of the sample. Over the survey period, 69 % of households had reported having other means of income outside farming, and the remaining 31 % depend solely on agriculture for their livelihood. While 81 - 85 % of households had access to agricultural extension and government - owned health services, 9 3 % ha ve access to a market for agricultural produce. Moreover, 56 % and 69 % of the households in the study reported to own mobile phones and radio, respectively. The average value of household assets that a typical household owns was around 14, 091 thousand Ugan dan Shillings (UGX 16 ,). Based on 2009 price index, the average price per Kilogram ( k g) of most commonly grown crops, namely beans, groundnuts, and wheat, was 982 , 993 and 454 UGX, respectively. Note however, that this is a snapshot picture of an average pri ce. There is a high variability in price between seasons, years, and locations , which is not captured here . 16 1 USD =1937.7 (2009); 1 USD =2481.5 (2014) Source; http s://www.calcprofi.com/exchange - rate - history - us - dollar - to - uganda - shilling.html 34 Related to geographical information, Uganda has four agroecological zones namely Tropic - warm/humid, Tropic warm/sub - humid, Tropic - cool/sub - humid an d Tropic - cool/humid with 5 4 %, 3 %, 13 %, and 3 0 % of the households living in each zone, respectively. The country has both bimodal and unimodal growing seasons. In the sample survey data, 69% and 31% of households have unimodal and bimodal growing seasons, r espectively. The country has an annual mean rainfall of 1195 millimeters (mm) and a mean temperature of around 22 degree Celsius . 35 Table 1 : Summary Statistics for Whole Sample and for the Final Y ear ( 2013/14 ) Variable N ame Description Mean Std. Dev. Mean Std. Dev. All Years (N= 6,489 ) 2013/14 (N= 1,118 ) Household level Food Security Outcome Variables HDDS Household Dietary Diversity Score (0 - 12) 7.52 2.19 8.27 1.90 MAHFP NF Number of food secure months in a year (0 - 12) 10.91 2.02 11.38 1.44 Net crop income Net crop income/day/HH (UGX,2009/10=100) 2672.38 3926.99 1981.88 2273.68 Gross crop income Gross crop income/day/HH (UGX,2009/10=100) 2887.11 4032.72 2130.54 2327.55 Calorie per capita per day Calorie produced per capita per day (Calories) 2258.02 3075.21 1669.53 1822.12 Protein per capita per day Protein produced per capita per day (Grams) 65.45 91.60 52.41 60.51 Explanatory Variables OR Treatment Variables *Cereal - legume inter =1 if the household adopt cereal - leg ume intercropping 0.52 0.50 0.51 0.50 *Cereal - legume rotation =1 if the h ousehold rotate cereals and legumes 0.42 0.49 0.42 0.49 *Legume - non - cer - inter =1 if the household adopt legume - non - cereal intercropping 0.50 0.50 0.46 0.50 *Legume - non - cer - rota =1 if the household adopt legume - non - cereal rotation 0.60 0.49 0.54 0.50 *Legume Monocropping =1 if the h ousehold adopt legume monocropping 0.47 0.50 0.46 0.50 *Cer - leg - intr area Cereal - legume intercropped area per HH (Acres) 0.58 1.01 0.51 0.74 *Cereal - leg - rota Cereal legume rotated area per HH (Acres) 0.68 1.22 0.61 0.97 *Legume - non - cer inter Legume non - cereal rotate intercropped area per HH (Acres) 0.59 1.03 0.40 0.61 * Legume - non - cer - rot Legume non - cereal rotated area per HH (Acres) 1.10 1.57 0.78 1.02 *Legume - mono - cropped Legume mono - cropped area per HH (Acres) 0.43 0.88 0.44 0.83 Control Variables Household Size Household size 6.11 2.95 6.28 2.99 Head Age Age of household head 48.14 15.14 50.25 14.90 Head Gender 1= Male, 0 otherwise 0.71 0.45 0.69 0.46 Head Edu. Education level household head (Years) 5.27 3.70 5.33 3.78 Total Area Planted Total area planted (Acres) 4.79 4.64 4.05 3.39 No. of Plots Number of plots per household 3.31 1.84 3.71 1.89 TLU Total tropical livestock units owned 2.07 10.14 1.74 4.68 Household Assets Total value of household assets ('000 UGX,2009/10=100) 14091.52 72389.53 5967.59 19752.66 Other Income Yes=1, if the household earns i ncome from other sources other than subsistence farming 0.69 0.46 0.41 0.49 Mobile Phone Yes=1, if the household owns Mobile phone 0.56 0.50 0.70 0.46 Radio Yes=1, if the household owns Radio 0.69 0.46 0.72 0.45 36 Table 1 Variable N ame Description Mean Std. Dev Mean Std. Dev Groundnut Price Previous season ave. groundnut price at district level (UGX/kg, 2009/10=100) 993.55 588.23 880.74 331.71 Beans Price Previous season ave. beans price at district level (UGX/kg, 2009/10=100) 982.79 642.60 752.96 148.38 Maize Price Previous season ave. maize price at district level (UGX/kg, 2009/10=100) 454.71 471.91 345.84 44.68 Urban 0 = Rural, 1 Urban 0.12 0.32 0.14 0.35 Community L evel Variables Yes=1, if g overnment clinic exists in the community 0.85 0.35 0.84 0.36 Agricultural Extension Yes=1, if a gricultural extension center exists 0.81 0.39 0.83 0.37 Market Yes=1 if the household has access to market, 0 otherwise 0.93 0.26 0.96 0.18 Dist. to Road HH distance to nearest major road (Kms) 8.42 7.43 8.85 7.67 Dist. to Market HH distance to nearest market (Kms) 32.56 18.68 33.14 18.86 Geographical V ariables Average Temp. Annual mean temperature ( o C * 10) 218.77 18.14 218.62 17.21 Annual Rainfall 12 - month total rainfall (mm) in Jan - Dec 1195.61 173.87 1196.35 139.55 Subsample of households with children All Years (N= 3,490 ) 2013/14 (N= 344 ) Child Level Nutrition Outcome Variables HAZ Height for age z - score - 1.49 1.68 - 1.32 1.67 WAZ Weight for age z - score - 0.75 1.22 - 0.60 1.27 WHZ Weight for height z - score 0.13 1.33 0.27 1.66 Stunting Yes=1, if the child HAZ z - score is < - 2SD , 0 otherwise 0.35 0.48 0.31 0.46 Underweight Yes=1, if the child WAZ z - score is < - 2SD , 0 otherwise 0.12 0.33 0.08 0.26 Wasting Yes=1, if the child WHZ z - score is < - 2SD , 0 otherwise 0.04 0.19 0.02 0.13 Child age Age of child (5 - 60 months) 33.50 14.62 41.00 10.98 Child gender =1 if the child gender is Male,0 if Female 0.49 0.50 0.43 0.50 Diarrhea Yes=1, if any child has d iarrhea during the past days 0.09 0.28 0.09 0.29 Number of Children Number of c hildren between ages 6 to 59 months/HH 1.78 0.75 1.73 0.86 Safe Drinking Water Yes=1, if the household has a safe means to drinking water 0.69 0.46 0.63 0.48 VIP Flush Latrine Yes=1, if the household has VIP or f lush toilet 0.02 0.15 0.01 0.12 Covered Pit Latrine Yes=1, if the household has covered pit toilet 0.75 0.44 0.72 0.45 Note: * represents the treatment variables. W hole sample (N= 6,489 ) . MAHFP NF (N=4,717). Subsample of households with children ( N =3,490). Net/Gross crop income are expressed Uganda shillings per day (UGX). Calorie/Protein produced per capit a per day are expressed in c alorie/grams. (1 USD =1937.7 (2009); 1 USD =2481.5 (2014). Source; https://www.calcprofi.com/exchange - rate - history - us - dollar - to - uganda - shilling.html . calculation based on LSMS - IAS data 37 The summary statistics related to child nutrition outcomes are from a subsample of households with children aged 6 to 59 months. From this sub sample, I took households with children in at least two rounds. Table 2 , represents the summary stati sti cs on stunting, underweight and w asting for children appearing in at least two rounds. Based on that, th e percentage of stunted , underweight and wasted children is 31 . 1 %, 7.7 % and 1.8 % during the last wave of the survey respectively. The prevalence of st unting , underweight and wasting has d o not show much variation across the four survey rounds, although there is a slight downward trend in these indicators over the five years between the first survey round (2009) and the last (2013/14) . This could potenti ally be due to the difference in proportion of children appearing in each wave, with the largest number of children in wave one and smallest in wave four ( Table 2 ). Table 2 : Prevalence of Stunting, Underweight and Wasting for Subsample of Households With 6 - 59 Months Children Child Nutrition Outcomes Proportion of children stunted, underweight and wasted at child and household level in each round 2009/10 2010/11 2011/12 2013/14 Total Sum Stunting 422(36.19%) 376 (35.84 %) 329(35.34%) 107(31.10 %) 1234(35.36%) Underweight 180 (15.44%) 120 (11.44 %) 103(11.06%) 26(7.56%) 429(12.29%) Wasting 59(5.06%) 36(3.44%) 26(2.81 %) 6(1.75%) 127(3.64%) No. of children 1166(33.41%) 1049(30.06%) 931(26.68%) 344(9.865) 3,490 No. of households 754(31.64%) 710(29.79%) 632(26.52%) 287(12.04%) 2,383 Note: Child level observations (N=3,490) on sub - sample of households with children (N= 2,383 ) of four survey years (2009 - 2014 ). The value in parenthesis represent proportion of children. Source; calculation based on LSMS - IAS data 4.2. Importance of Different Legume - B ased Cropping in the Ugandan Context In Uganda, as shown in Figure 3 , the proportion of households that grew legumes was in the range of 86 - 90 % during the survey period. Among all legume - based cropping , (i.e. , use of each legume - based cropping is not mutually exclusive and each household can adopt multiple practices at spe cific season or year) . legume - noncereal - rotation was dominant followed by cereal - legume intercropping , legume - noncereal intercropping legume monocropping, and cereal - legume rotation in descending order of importance . The percentage of households rotating legume s with non - cereal 38 crops was highest in 2010 ( 63 %), followed by years 2009/10 and 2011/12 ( 60 %), and 2013/14 (54% ) , while cereal - legume rotation prevalence was the smallest among all practices (41 - 43%. The proport ion of households that intercrop legumes with cereals/noncereal remained indifferent ranging from 51 to 52% and that of legume monocropping households slightly varies among the survey years (50% in 2011/12 and 46% in 2013/14). Overall, there is not much di fference in the overall proportion of households that integrate legumes in their cropping systems across the survey years, but the frequency of use of different practices varies from year to year (see Table 3 ). Figure 3 : Proportion of Households Using Different Legume - Based Cropping in Uganda Note: Household level observations (N= 6,489 ) of four survey years (2009 - 2014). Legume based cropping per each household are not mutually exclusive. Source; calculation based on LSMS - IAS data As shown in Table 3 below , households practice legume - non - cereal rotation ( 10.9 %) more sequentially than cereal - legume intercropping ( 10.6 %), legume - non - cereal intercropping ( 9.8%) legume monocropping ( 8.4 %), and cereal - legume rotation ( 4 .1%), indicating a significant intra - household variation of technology adoption over the survey years. Similarly, the most adopted technology was legume - non - cereal rotation, while 25 . 1 % of the household s did not rotate cer eals and legumes within the survey period. These variability across time (years) supports using the 86 86 90 86 60 63 60 54 52 51 52 51 50 50 52 46 43 48 50 46 43 41 41 42 0 20 40 60 80 100 2009/10 2010/11 2011/12 2013/14 Proportion of Households Using Different Legume - Based Cropping in Uganda Legumes Legume non-cereal rotation Cereal-legume intercropping Legume non-cereal intercropping Legume monocropping Cereal-legume rotation 39 household fixed effect model approach as an identification strategy to estimate the causal impacts legume - based cropping. Table 3 : Frequency of Legume - Based Cropping Adoption a cross the Four Survey Rounds . Type of Technology Number of survey roun ds in which a household reported using a given practice (out of total 4 rounds) 0 1 2 3 4 None Round Rounds Rounds Rounds Total Legume - non - cereal rotation (%) 12.1 22.2 28.7 26.0 10.9 100 Cereal - legume intercropping (%) 20.9 22.1 24.4 22.1 10.6 100 Legume - non - cereal intercropping (%) 20.7 24.8 25.1 19.7 9.8 100 Legume mono - cropping (%) 21.5 26.5 24.8 18.8 8.4 100 Cereal - legume rotation (%) 25.1 31.0 25.8 14.1 4.1 100 Note: Household level observations (N= 6,489 ) of four survey years (2009 - 2014). calculation based on LSMS - IAS data Before discussing the empirical results, I present the results of a two - tailed mean test for the main outcome variables with respect to users and non - users of each legume - based cropping . As shown in Table 4 , m ean comparisons (using t - test) of outcome indicators between users and non - users suggest that users of legume - bas ed cropping are better - off in most of the food security outcome variables ; households that used cereal - legume intercropping, cereal - legume rotation, legume monocropping, legume - non - cereal intercropping and legume - non - cereal rotation, report higher calorie and protein production, net crop income, HDDS, and MAHFP. Nonetheless the difference is weaker down the path; I did not find much difference on child nutrition outcome indicators between users and nonusers of each legume - based cropping. Even though I found positive correlation between legume - based cropping and the outcomes suggest causality. Hence the next section will explore further thi s issue of causality using econometric approaches . 40 Table 4 : Comparison of Means of all Dependent Variables, a mong Users and Non - users of Different Legume - B ased C ropping Outcome Variables Cereal legume intercropping Cereal legume rotation Legume - non - cereal intercropping Legume - non - cereal rotation Legume monocropping Adoption Decision No Yes P - val. No Yes P - val. No Yes P - val. No Yes P - val. No Yes P - val HDDS 7.3 7.7 *** 7.5 7.6 ** 7.4 7.6 ** 7.4 7.6 *** 7.5 7.5 -- MAHFP F 10.8 11.0 *** 10.9 10.9 -- 10.9 10.9 -- 10.9 10.9 -- 10.9 10.9 -- Net crop income 2426.9 2902.8 *** 2496.7 2917.1 *** 2344.5 3002.0 *** 2088.5 3065.2 *** 2268.9 3125.1 *** Gross crop income 2635.6 3123.2 *** 2693.8 3156.4 *** 2557.3 3218.7 *** 2287.3 3290.7 *** 2440.7 3388.0 *** Calorie per capita per day 2041.6 2461.2 *** 2043.5 2556.7 *** 2212.4 2303.8 -- 2013.3 2422.7 *** 1981.2 2568.6 *** Prot ein per capita per day 56.6 73.8 *** 56.9 77.4 *** 63.2 67.7 ** 56.8 71.3 *** 54.0 78.3 *** HAZ - 1.46 - 1.53 -- - 1.48 - 1.51 -- - 1.49 - 1.50 -- - 1.43 - 1.54 ** - 1.51 - 1.47 -- WAZ - 0.74 - 0.75 -- - 0.76 - 0.73 -- - 0.75 - 0.75 -- - 0.72 - 0.77 -- - 0.79 - 0.70 *** WHZ 0.11 0.15 -- 0.10 0.18 * 0.12 0.14 -- 0.12 0.14 -- 0.09 0.19 ** Stunting 0.33 0.37 ** 0.35 0.36 -- 0.34 0.36 -- 0.34 0.36 * 0.37 0.34 ** Underweight 0.12 0.13 -- 0.12 0.13 -- 0.12 0.13 -- 0.12 0.12 -- 0.14 0.11 *** Wasting 0.04 0.04 -- 0.03 0.04 -- 0.04 0.04 -- 0.04 0.04 -- 0.04 0.03 *** Notes: Yes= adoption and No=non - ado ption of each legume - based cropping . P - Val. Indicates P value of mean test . Household level observations (N= 6,489 ), MAHFP F observations (N=4,717), Child level observations (N=3,490) of four survey years (2009 - 2014). (***, **, *, and -- represents statistically significant at 1, 5 and 10% and nonsignificant values of a two - tailed test respectively). HDDS and MAHFP values range between 0 to 12. Net/Gross crop income are expressed in Uganda shillings per day (UGX). Calorie/Protein pr oduced per capita per day are expressed in Calorie/grams. HAZ, WAZ and WHZ represent raw z - scores. Stunting, Underweight and Wasting takes either 1 or 0 values. (1 USD =1937.7 (2009); 1 USD =2481.5 (2014). Source; https://www.calcprofi.com/exchange - rate - history - us - dollar - to - uganda - shilling.html Source; calculation based on LSMS - IAS data 41 4.3. Impact of Legume - Based Cropping on Household level Food Security and Child Level Nutrition Outcomes To assess the impact of each of the legume - based cropping on food security and nutrition outcomes (household welfare), I applied all the regression models explained in Chapter 3 on the un balanced household panel data (N= 6,489 ). Tables 5 to 9 show the summary of regression results for each legume - based cropping with respect to each outcome variable. Each column in Tables 5 to 9 represents different estimators used. To capture the impact of the prevalence and extent of use of a given legume - based cropping , both binary and continuous treatment variables were used. The full regression results including the coefficients of all control variables are shown in Appendix B ; Tables B 1 - B 7 . In this study, I used e ight different outcome variables (representing different dimensions of food security), five main treatment varia bles (i.e., legume - based cropping ), and a set of control variables. The POLS and fixed effect estimator coefficients are reported as average effects . Before I discuss the main results of the model estimation for treatment variables (i.e., results in Tables 5 - 9 ), I briefly discuss the results related to the effect of control variables on the food security indicators reported in Tables B1 - B 2 in A ppendix B . Based on POLS regression results, among the main variables representing household characteristics, household size has a negatively significant effect on production outcome variables except for net crop income, reflecting the higher probability of food insecurity with increased family size . Gender of household head has significant effect on many of the outcome variables, indicating male - headed households are more productive and food secure than female - headed households in Uganda . This is consistent with the findings of Smith et al. (2006) ; with the exception of Mozambique, female - headed hou seholds were less productive in many SSA countries they studied, including Uganda. Both a ge of household head remained in significant in all of the outcome variables. Similarly, education of household head has shown a positive and signifi cant effect on many remained 42 insignificant, i mplying a nonlinear relationship. Not surprisingly, variables that represent assets owned by each household, namely total area pl anted, and number of plots owned has shown a positive and significant impact on many of production outcomes, indicating the positive contributions of asset ownership to enhancing food security. Non - farm assets like radio and mobile phone ownership, which i s a proxy for both wealth and access to information, has a positive and significant effect on most of the outcome variables. Surprisingly, access to other forms of income, other than crop income, has remained insignificant on most of food security outcome s, except on HDDS. The results on HDDS suggest that households with non - farm sources of income are likely to have eaten more diverse types of foods in the past 7 days than households with no non - farm sources of income. nd significant impact on net crop income indicating, that those living in urban areas sell more crops than those living in rural areas, as they might have more access to markets. Moreover, t he variable urban has shown positive and significant effect on MAHFP and HDDS, implying urban dwellers might have more access to food compared to rural dwellers. As far as the round (time) effect is concerned, I foun d a heterogenous effect of time (i.e., y ear) on many of the outcome variables indicating time have had an effect throughout the survey rounds. Region - wise, there was a negatively significant difference between Central and Eastern , but that of Northern and Western effect was dropped from the ana lysis. Turning now to the main results, I discuss the effects of each of the treatment variable on the outcome indicators. 4.4. Effects of Cereal - Legume Intercropping on Food Security and Nutrition Related Welfare Indicators Due to the advantages of reaping multiple crops from the same plot in the same season, cereal - legume intercropping was may have a positively significant impact on food production, income, and nutrition outcomes. As reported in Table 5, I found positive and significant effect of cereal legume intercropping on calorie and protein produced, but effect is not robust across all 43 estimators used. The effect remained insignificant on many of the food security and nutrition outcomes namely crop income, HDDS, and MAH FP, The POLS and FE regression results suggest that if a household adopts cereal - legume intercropping, on average it reaps additional 259 - 428 calories and 11 - 19 grams of protein per capita per day more than the household that do not practice cereal - legume intercropping, keeping all other factors constant (Table 5) . Similarly, a houshold that alloted an additional acrea of land to cereal - legume intercropping, on average reaps 8 grams of addtiional protein per capita per day relative to nonusers, ceteris pari bus . These results indicate that relative to the mean average on calorie/protein (2258 calories/65.4 grams of protein for all houeholds, the increase in calorie and protein approximates to about 15 - 18%/16 - 29% in the case of binary and continous treatment variables, respectively. Even though, I expected positive impact of cereal legume intercropping on HDDS the results turned out to be insignificant. As a reminder, the HDDS indicator used in this study was estimated using 7 - days recall period and not 24 - ho ur recall. Hence, the longer time frame might have inflated the diversity score and thus resulted in biased estimates, as households are likely to consume and report more variety of foods in seven days than in 24 hours. A study by Kibrom and Qaim (2016) use d seven days recall data for HDDS and found a significant impact of production diversity on that indicator in Uganda. Turing to the analysis downstream in the agriculture - nutrition linkages pathway, the fixed effects results of cereal - legume intercropping on child nutrition outcomes has shown negative and significant impact of cereal - legume intercropping on HAZ z - score, i ndicating that an additional land allotted to cereal legume intercropping can potentially decrease - score by 0.058 points, keeping all other factors constant. Similarly, using POLS estimator the coefficient on underweight is negative and sig nificant indicating that an additional land allotted to this practice reduce the probability of wasting by 0.57 percent, ceteris paribus. Compared to the mean average on 44 HAZ/wasting ( - 1.49/0.04), the effect of this practice on HAZ and wasting is not substa ntial (4% and 14% of the mean HAZ z - score and wasting). However given that the POLS does not account for unobserved heterogeneity bias, results based on POLS need to be interpretted with caution. Overall, the effect of cereal - legume intercropping on food s ecurity indicators has shown little effect but remained insignificant in many of the outcome variables. The findings on this study are consistent with the findings by Sauer et al. (2016) for Zambia, where they also found little effect or insignificant effe cts of cereal - legume intercropping on cereal growing households. 45 Table 5 : Main Regression Results related to the E ffects of Cereal - Legume Intercropping on Household Level Food Security and Child Level Nutrition Outcomes Treatment Variable s Binary (=1 if HH intercrops cereals & legume crops ) Continuous (Total acres of cereal - legume intercropped) Estimator POLS Coef FE Coef POLS Coef FE Coef Outcome Variables Household Level O utcomes Calorie per capita per day (Calories) 427.608*** 258.881* 203.683 88.493 (127.986) (140.725) (138.955) (150.152) Protein per capita per day (Grams) 18.844*** 11.091*** 7.671** 2.034 (3.973) (3.435) (3.556) (3.778) Gross crop income 7.391 128.099 - 109.259 20.228 (UGX,2009/10=100) (140.357) (162.628) (93.814) (115.238) Net crop income 11.117 121.871 - 117.303 14.201 (UGX,2009/10=100) (137.113) (159.134) (92.288) (113.957) HDDS 0.001 0.014 0.038 0.029 (0.084) (0.082) (0.049) (0.043) MAHFP NF 0.031 0.090 - 0.054 - 0.023 (0.073) (0.101) (0.045) (0.055) Child L evel Outcomes HAZ (Height for age z - score) - 0.060 - 0.035 - 0.006 - 0.058* (0.072) (0.074) (0.036) (0.034) WAZ (Weight for age z - score) - 0.049 - 0.017 0.004 - 0.008 (0.057) (0.059) (0.025) (0.024) WHZ (Weight for height z - score) - 0.008 0.022 0.005 0.029 (0.063) (0.078) (0.029) (0.027) Stunting (HAZ z - score is < - 2SD) 0.0328 0.0230 0.0060 0.0131 (0.0215) (0.0224) (0.0101) (0.0107) Underweight (WAZ z - score is < - 2SD) 0.0090 0.0012 - 0.0067 - 0.0016 (0.0154) (0.0162) (0.0064) (0.0065) Wasting (WHZ z - score is < - 2SD) - 0.0033 - 0.0063 - 0.0057* - 0.0038 (0.0078) (0.0110) (0.0032) (0.0043) Note: (*** p<0.01, ** p<0.05, * p<0.1, robust standard errors are in parenthesis); Household level observations (N= 6,489 ), MAHFP F observations (N=4,717), Child level observations (N=3,490) of four survey years (2009 - 2014). POLS and FE standard errors are clustered at household level and robust to serial correlation and heteroskedasticity . HDDS and MAHFP values range between 0 to 12. Net/Gross crop income ar e expressed in Uganda shillings per day (UGX). Calorie/Protein produced per capita per day are e xpressed in c alorie s /grams. HAZ, WAZ and WHZ represent raw z - scores. Stunting, Underweight and Wasting takes either 1 or 0 values. (1 USD =1937.7 (2009); 1 USD =2481.5 (2014). Source; https://www.calcprofi.com/exchange - rate - history - us - dollar - to - uganda - shilling.html Source; calculation based on LSMS - I AS da ta 46 4.5. Effects of Cereal - Legume Rotation o n Food Security and Nutrition Related Welfare Indicators The presentation of regression results in Table 6 for cereal - legume rotation follows similar format as cereal - legume intercropping regression results, where each column represents the estimator used, and the rows represent the outcome variables. Overall, c ereal - legume rotation results are weaker than cereal legume intercropping and remained insignificant on most of the outcome variables except on protein produced, WAZ and WHZ. But even for these few outcomes, results are not robust across all estimators use d. Results of the POLS model suggest that keeping all other factors constant, households that rotate legumes with cereals on average gain 8 grams of more protein per capita per day than the households that do not practice cereal - legume rotation; indicating an average increase of about 12% relative to the average protein production per day of 65 g in the sampled households. FE effect results suggest that one acre of land rotated with cereals and legume on average increases the WAZ and WHZ z - scores by 0.08 and 0 .03 points, respectively, keeping all other factors constant. Nonetheless these results are not robust across all estimators and type of child nutrition outcome indicators and need to be interpreted with caution. Overall, my analysis suggests little effec t of cereal - legume rotation on the immediate food production indicators measured by protein produced per capita per day, and child nutrition outcomes namely WAZ and WHZ z - scores, but the effect remained insignificant on all other variables. 47 Table 6 : Main Regression Results related to the Effects of Cereal - Legume Rotation on Household Level Food Security and Child Level Nutrition Outcomes Treatment Variable s Binary (=1 if HH rotates cereal & legume crops ) Continuous (Total acres of cereal legume rotated) Estimator POLS Coef FE Coef POLS Coef FE Coef Outcome Variable s Household Level O utcomes Calorie per capita per day (Calories) 205.268 115.470 - 4.583 - 51.337 (139.287) (134.319) (90.913) (96.834) Protein per capita per day (Grams) 7.955* 5.026 1.050 - 0.531 (4.280) (3.358) (2.439) (2.533) Gross crop income - 145.913 - 86.536 - 42.793 - 68.478 (UGX,2009/10=100) (133.256) (142.007) (76.364) (77.651) Net crop income (UGX,2009/10=100) - 171.552 - 115.746 - 55.930 - 84.871 (131.684) (140.461) (75.582) (77.151) HDDS 0.077 0.047 0.012 - 0.017 (0.080) (0.079) (0.041) (0.036) MAHFP NF - 0.122 - 0.151 0.008 - 0.011 (0.075) (0.092) (0.030) (0.036) Child Level O utcomes HAZ (Height for age z - score) - 0.016 0.030 0.005 - 0.003 (0.070) (0.065) (0.031) (0.030) WAZ (Weight for age z - score) 0.049 0.076* 0.016 0.028 (0.054) (0.043) (0.021) (0.019) WHZ (Weight for height z - score) 0.069 0.054 0.020 0.034* (0.058) (0.054) (0.019) (0.020) Stunting (HAZ z - score is < - 2SD) - 0.0004 - 0.0129 - 0.0022 - 0.0016 (0.0205) (0.0215) (0.0085) (0.0081) Underweight (WAZ z - score is < - 2SD) - 0.0015 - 0.0179 0.0027 - 0.0031 (0.0144) (0.0138) (0.0054) (0.0054) Wasting (WHZ z - score is < - 2SD) 0.0074 0.0039 0.0026 - 0.0011 (0.0076) (0.0090) (0.0026) (0.0037) Note: (*** p<0.01, ** p<0.05, * p<0.1, robust standard errors are in parenthesis); Household level observations (N= 6,489 ), MAHFP F observations (N=4,717), Child level observations (N=3,490) of four survey years (2009 - 2014). POLS and FE standard errors are clustered at household level and robust to serial correlation and heteroskedasticity. HDDS and MAHFP values range between 0 to 12. Net/Gross crop income ar e expressed in Uganda shillings per day (UGX). Calorie/Protein produced per c apita per day are expressed in c alories/grams. HAZ, WAZ and WHZ represent raw z - scores. Stunting, Underweight and Wasting takes either 1 or 0 values. (1 U SD =1937.7 (2009); 1 USD =2481.5 (2014). Source; https://www.calcprofi.com/exchange - rate - history - us - dollar - to - uganda - shilling.html Source; calculatio n based on LSMS - IAS dat a 48 4.6. Effects of Legume - Noncereal Intercropping on Foo d Security and Nutrition Related Welfare Indicators The summary of regression results on legume - non - cereal intercropping is presented in Table 7. As in previous tables, each column represents different estimators used. Unlike the results for cereal - legume intercropping and rotation, intercropping legumes with non - cereal crops has shown positive and significant (p<.05) effect for all categories of crop income, but sensitive to the type of treatment variable used. Based on the FE regression results for continuous treatment variables , an additional acre of land allotted to legume - non - cereal intercropping increases gross/net crop income by an average of 376/370 UGX per day, keeping the effects of all other factors constant. This effect is non - negligible (13%/14%) when compared to the daily gross/net crop income of 2887/2672 UGX. Non etheless, the effect of legume non - cereal intercropping on gross/net crop income remained insignificant in the case of binary treatment variables. The study however did not find any effect of legume non - cereal intercropping on the remaining production outc ome variables (calorie and protein produced). Regarding consumption related outcomes, I found positive/negative and significant (.05